Physics Context Builders: Modular Simulation Frameworks

• Physics Context Builders (PCBs) are modular frameworks that automate the assembly of simulation environments by externalizing geometry, models, and experiment configurations.
• They enable rapid prototyping and reproducible research across diverse fields such as particle physics, computational modeling, and vision-language integration.
• PCBs improve accuracy and flexibility by decoupling context definitions from static code, as demonstrated in applications like GEARS, TurboPy, and SARAH.

Physics Context Builders (PCBs) are a class of frameworks, algorithms, and methodologies that automate the assembly and management of the “physics context” required for high-fidelity modeling, simulation, or reasoning about physical systems. The core concept is to externalize or modularize the specification and orchestration of geometry, fields, experiment parameters, and data‑handling strategies—eliminating hardwired dependencies on specific codebases or models. PCBs span applications in particle physics simulation, condensed matter and quantum device characterization, computational modeling, automated model building in high energy physics, and, more recently, vision-LLM reasoning and in-context learning with LLMs. They support reproducibility, rapid prototyping, and transferability by enabling every domain-specific aspect to be specified, inspected, and manipulated at runtime, or learned and composed dynamically.

1. Defining Characteristics and Scope

A Physics Context Builder (PCB) is defined as a framework, toolchain, or modular subsystem that orchestrates all components necessary for the definition and execution of a physics-based computation, experiment, or reasoning task—typically at runtime, via configuration, simulation, or in-context learning—without requiring alteration or recompilation of underlying implementation code (Liu, 10 Dec 2025, Richardson et al., 2020, Balazadeh et al., 2024). This includes, but is not limited to:

• Geometric specification: full definition of spatial domain, material properties, boundary and interface conditions.
• Model and process selection: choice and parameterization of physical processes, physics lists, and interaction mechanisms.
• Source/initial condition control: specification of primary sources (particle beams, fields, stimuli), often through rich scripting or macro interfaces.
• Data acquisition and diagnostics: definition of diagnostics, output formats, and analysis pipelines, typically in standardized formats.
• Separation of runtime configuration from implementation: users define or manipulate all details through external configuration files, macro commands, or high-level APIs; the underlying code remains static.

PCBs are distinguished from monolithic frameworks or hardwired “application codes” by their modularity, explicit configuration surface, and focus on rapid, reproducible customization.

2. Exemplary Realizations in Simulation and Modeling

PCBs are instantiated in a variety of computational and experimental domains, each reflecting the paradigm’s key principles.

2.1 Particle Physics Simulation: GEARS

GEARS is a prototypical PCB for the Geant4 simulation toolkit. Its approach is characterized by:

• Geometry definition: supports both GDML (XML-based) and Markdown-inspired .tg formats, parameterized by text files; these can express repeated structures and complex transformations.
• Physics list selection: leverages the PHYSLIST factory via environment variables, with post-selection fine-tuning through macro commands, all external to the C++ code.
• Primary source definition: delegates to the General Particle Source (GPS) macro system, supporting fine-grained, runtime-directed control over source distribution and kinematics.
• Data output: uses a flat, branch-minimized ntuple structure for high-speed ROOT/TTree analysis, eschewing nested C++ objects.
• Zero-step capture: implements tracking via G4SteppingVerbose to record state at event initiation, not accessible in the standard SteppingAction (Liu, 10 Dec 2025).

The entire GEARS application is a static executable; iteration is achieved solely by editing text files, environment variables, and macro command scripts.

2.2 Modular Computational Physics: TurboPy

TurboPy embodies the PCB paradigm in Python for rapid prototyping of grid-based and particle-in-cell (PIC) simulations:

• Class-based decoupling: a Simulation object owns and wires together PhysicsModule, Grid, Clock, ComputeTool, and Diagnostic components.
• Publish–inspect pattern: modules declare and exchange resources via publish_resource and inspect_resource interfaces to ensure loose coupling.
• Configuration-driven instantiation: setups are specified in TOML files or Python dicts, enabling reproducibility and rapid reconfiguration without modifying source files.
• Diagnostics and workflow orchestration: the Simulation manages stepping, diagnostics, and state handoff among modules in a controlled, modular fashion (Richardson et al., 2020).

2.3 Automated Model Construction: SARAH 4

SARAH 4 in high energy physics automates the context building for BSM (beyond-standard-model) computations:

• Model input via concise semantic files: gauge groups, matter content, Lagrangian, and symmetries are specified in Mathematica lists.
• Algebraic derivation: the entire chain from Lagrangian to vertices, mass matrices, tadpoles, RGEs (up to two-loop), and one-loop self-energies is generated automatically.
• Cross-tool interoperability: outputs are generated for FeynArts, Calchep, CompHEP, WHIZARD, UFO, Vevacious, and SPheno, supporting full-chain automation from symbolic computation to Monte Carlo event generation (Staub, 2013).

3. PCBs in Vision-Language and Data-Driven Physical Reasoning

The PCB concept extends to modern AI, notably in physical reasoning for vision-LLMs (VLMs):

• Context generation: small, specialized VLMs (context builders) are fine-tuned to output physical scene descriptions (objects, dynamics, stability) from images or video.
• Compositional modularity: context builders are trained with simulation-driven ground truth and deliver scene summaries consumed by large, generalist LLMs via in-context prompting.
• Modular inference workflow: a triage agent routes inputs to the relevant context builder, which generates the context fed to the LLM for downstream reasoning. This decouples perception and physical interpretation from high-level reasoning (Balazadeh et al., 2024).

Empirical results demonstrate ∼15–25% accuracy improvements on complex physical reasoning tasks and strong Sim2Real transfer, validating the modular PCB approach in AI physical reasoning.

4. PCB Patterns in In-Context Learning

PCBs can be realized in a latent, implicit sense inside LLMs:

• Emergent representation discovery: LLMs prompted with raw time-series data of physical variables internalize conserved quantities (e.g., total energy, kinetic, and potential energy) in their residual activations.
• Sparse autoencoder analysis: applying SAEs to activation streams reveals dimensions with high Pearson correlation to ground-truth physical invariants. These appear and peak in mid-layers of the transformer.
• Functional necessity: ablation experiments confirm that removing the leading “energy code” activations significantly degrades forecasting performance, demonstrating that LLMs construct implicit physics contexts in-flight (Song et al., 17 Aug 2025).

A plausible implication is that prompt engineering and dictionary code extraction could be harnessed to expose and interface with these emergent latent PCBs.

5. Experimental PCB Paradigms in Cryogenic Quantum Devices

PCBs are employed experimentally in the characterization and calibration of quantum device interconnects:

• Self-calibrating testbeds: a λ/2 open-circuit coplanar waveguide resonator is used to measure temperature-dependent dielectric properties (real permittivity εᵣ(T) and loss tangent tan δ(T)) of interface PCBs.
• Physics context as system state: resonator geometry, environmental parameters, and simulation-informed calibration curves constitute the full context; measurements guide adaptation of design parameters for quantum-coherent applications (Paghi et al., 2023).
• Design guideline codification: the PCB doubles as a sensor, integrating physical measurement, simulation correction, and actionable output for device engineering.

6. Limitations, Extensibility, and Best Practices

Key limitations and extension points are domain-specific:

• Dimensionality and scalability: some PCBs (e.g., TurboPy) are currently limited to 1D, requiring extension for multidimensional, AMR, or advanced solver support (Richardson et al., 2020).
• Interoperability: cross-tool output support—exemplified by SARAH—requires precise mapping to heterogeneous toolchains, and maintenance as formats evolve.
• Error propagation and accuracy: manufacturing or configuration tolerances in experimental PCBs (e.g., sub-200 μm features in interface PCBs) directly affect physical interpretation accuracy (Paghi et al., 2023).
• User interface: PCB modularity places a premium on robust, expressive configuration and domain-specific languages.
• Automation versus interpretability: latent PCBs in LLMs are difficult to inspect; further research is needed to surface and leverage their internal context construction for interpretability and control (Song et al., 17 Aug 2025).

Recommended best practices include modularizing all computational components, adopting standardized output formats, incorporating diagnostic feedback, and balancing configurability with system stability.

7. Significance and Impact across Fields

PCBs have transformed both computational and experimental workflows:

• Simulation flexibility: enabling iterative, reproducible simulation and analysis cycles without recompilation or low-level code edits (Liu, 10 Dec 2025).
• Reproducibility and portability: ensuring that full physics contexts are documented and transferable across platforms and collaborators (Staub, 2013, Richardson et al., 2020).
• AI interpretability: new insight into how neural models encode and reason about physics in context, opening prospects for dynamical system inference and improved physical robustness (Balazadeh et al., 2024, Song et al., 17 Aug 2025).
• Experimental design and validation: streamlined protocols for device characterization, fostering rapid hypothesis testing and feedback-driven optimization in hardware (Paghi et al., 2023).

As both explicit toolchains and emergent algorithmic motifs, PCBs provide a scalable abstraction for organizing, manipulating, and interrogating the complex context needed for modern physical science and engineering.